When is behavior?

[From Bill Powers (961128.2200 MST)]

Bruce Abbott (961127.2010 EST)

Unless I just _changed_ my posture, it is not relevant, not an act.

Tracy Harms (1996;11,27.19:12 MST)

This looks very wrong to me. There are perceptions being actively
controlled by a given posture even when that posture goes unchanged. Every
posture is the result of a combination of muscular tensions, and the nature
of muscles means that there are error-corrections which alter these muscle
tensions over time as a given posture is held *unchanged*.

This is a very important point. Tracy is right. When a control system is
actively controlling, it may not be doing anything that an external observer
can see as an "event." But it is actively controlling; the slightest
disturbance will result in opposition, the slightest change in the reference
signal will produce a corresponding change in the controlled variable. A
controlled variable doesn't actually have to vary to be either controlled or
a variable. The output of the control system is ALWAYS having consequences,
even when it's constant. The same control system, turned off, may exhibit
exactly the same observed absence of behavior, but it is not controlling.

Bruce says

I agree completely that we are actively controlling perceptions all the
time, just as you (and PCT) describe. But it is the consequences of
movements that I want the model to focus on -- the dynamic changes in the
values of these movement-related perceptions.

This forces the discussion back to the level of events instead of continuous
relationships. Remember my discussion of the lever. The position of the far
end of the lever is always related to the position of the near end by the
constant of proportionality. The position of the far end is ALWAYS a
consequence of the position of the near end, even when both ends are
stationary. You can create the picture of one event causing another event
simply by moving the near end in a way that you perceive as an event -- for
example, a quick wiggle up and down. This event will have the consequence
that the far end will move in a quick wiggle down and then up. But that has
nothing to do with the actual input-output relationship. The "eventness" of
this observation is simply an artifact of the way the independent variable
is "moved." A "movement" is simply a change in position seen as an event.
All during a movement of one end of the lever, the position of the other end
remains in the same relation to the end that is manipulated. It is this
continuous relationship that explains how input and output variables are
related -- not the particular way in which the input variable may change (or
not change).

When I took graduate courses in a Spencean department, it was emphasized
over and over that you had to set up experiments in a way that allowed
definite events to be observed. In a maze-running experiment, for example,
you can't just put the rat into the starting chamber and let it go. The rat
moves around in a continuous way, eventually becoming oriented and starting
to move up the alley. But this doesn't give you any definite moment to call
the "start" of the run. What you need to do is to use a photocell or a
swinging barrier that records the start as a relay or switch closure; this
gives you an event that occurs at a specific moment in time. Then you have
something to measure. The same goes for the end of the run; again you need a
relay or switch closure to create a specific event marking the end. Once you
have reduced behavior to events, you can count them, calculate their
probability of occurrance, measure the elapsed time between them, and
measure their frequency.

This requirement reflects the observer's assumption that environmental
events cause behavioral events. This is how Skinner thought of
"consequences." A consequence is an event produce by a preceding series of
behavioral events, marked by contact closures. The probability of a contact
closure can be related to the frequency of contact closures, and the theory
of behavior can be stated in terms of events and their effects on
probabilities or frequencies of events.

I think it's interesting and significant that in physics, events per se are
not considered causal. If you drop a steel ball on a steel plate, the
velocity of the ball reverses at the moment of contact, creating an event we
call an "elastic collision" or a "bounce." But it is not the "bounce" that
causes "reversal" of the velocity; that is just a convenient way to refer to
the observation in human terms. What causes the bounce is an episode of
elastic compression of the ball and the steel plate, which decelerates the
ball to a stop and then, as the compression continues to react against the
ball, accelerates the ball upward, while various momenta are conserved. The
event really consists of a series of ongoing smooth interactions that begin
imperceptibly as the fields surrounding the atoms of the ball approach the
fields surrounding the atoms of the steel plate, go through continuous
smooth changes, and finally die out again imperceptibly as the fields
separate again. The "eventness" of this interaction through time is entirely
in the eye of the beholder, because human beings (for whatever reasons) find
it convenient and natural to perceive in terms of events. And it is clearly
not the "eventness" that is causing the phenomenon; just the opposite. The
underlying continuous interactions through space and time --plus the
properties of human perceptions -- are what create the subjective
impressions of events.

This picture becomes confused when we take an animal that behaves
essentially continuously and put it into a situation where the _environment_
is set up to create events. As the rat approaches the bar, its body and paws
are moving smoothly and continuously through various patterns of change. It
noses around the bar and rears up, all in a continuous flow of movement. At
some point, the weight of the rat depresses the bar far enough to cause a
microswitch to snap from open to closed. At that instant, an "event" is
recorded. What the rat was doing immediately before and immediately after
this event is NOT recorded, except by the video camera if the experimenter
has provided one. As far as the data are concerned, everything that the rat
did was done at the instant that the microswitch snapped closed; before and
after that instant, the rat was doing nothing.

The apparatus, of course, is set up so that the input events are counted or
timed, and when a certain criterion is met, the output event is triggered; a
pellet drops suddenly into a food dish. At this moment, a "reinforcer" is
said to have been administered. The apparatus is set up to conform to the
experimenter's idea that natural phenomena occur as discrete events. The
events seems discrete, that is, on the scale of normal human perception.
They are all actually continuous processes, although they may happen more
rapidly than we can follow.

Now consider this reinforcing event from the standpoint of the rat's
continuous sensory experience of the world. The rat's nose and eyes are
always providing it with a picture of the world; as the rat moves around
this picture continuously changes. As the rat's nose passes close to the
food dish, the smell of the food in the magazine, and the flakes of food
adhering to the surface of the cup, becomes stronger. The rat zeros in on
this smell, but tasting and seeing no food there passes on to other
activities. Its world is full of continuously waxing and waning scents;
whiffs of air containing moisture, the smells of previous occupants of the
cage, smells of traces of urine and feces both its own and from others of
its species who have been in the cage. It sees the items in the cage; glints
of light, dark places, and so on -- who knows what a rat sees?

Into this continuously changing world of perception there comes a sudden
change: a piece of food falls into the cup, rattles and bounces for a time
that seems short to a human but probably seems longer to a rat, with its
short-coupled nervous system. The smell of food intensifies and the rat sees
the food in the cup. It moves itself to the cup, rearing up, twisting its
head to get its mouth into the cup, bringing its forepaws into play to seize
the food and move it from the cup into its mouth. It salivates and chews,
intensifying the tastes and smells of the food, and passing the food inward
to its stomach. While it's doing this, it may be balancing on its
hindquarters and turning the piece of food with its paws as it nibbles, or
it may move back to the lever to continue its explorations until another
contact closure takes place, chewing on the previous piece of food all the
while. It may move back to the cup, but have to wait until it has finished
chewing and swallowing the previous piece of food before it has room to
stuff the new one into its mouth.

At the moment that the food is released to fall into the cup, a
"reinforcement" is recorded. Everything that happens to the rat, data-wise,
is compressed into this brief moment. Nothing the rat was sensing in any
modality before or after this event is recorded.

So what we end up with is a data set consisting of instantaneous events
occuring at specific instants of time, with no record of all the continuous
processes of action and (inferred) perception that were going on between the
events.

If rats can indeed perceive and control in terms of events, defined as we
perceive them, then experiments like these might tell us something about how
they do this. But if they can't (or if they are nowhere near as good at this
as we may think), then we can still read event-based phenomena into the
results simply because we have forced the environment to work in terms of
discrete events rather than continuous variations.

In order to resolve this problem, the first thing we have to do is to
recognize events not as natural units of physical phenomena, but as modes of
perception. As long as we think that events are natural units, we will fail
to see that we are making a specific hypothesis about what the rats perceive
and what their internal mode of operation is. To recognize this we have to
see first how we ourselves impose eventness on an otherwise continuous world
of experience.

Best,

Bill P.

[From Bruce Abbott (961129.1440 EST)]

Bill Powers (961128.2200 MST) --

Bruce Abbott (961127.2010 EST)

I agree completely that we are actively controlling perceptions all the
time, just as you (and PCT) describe. But it is the consequences of
movements that I want the model to focus on -- the dynamic changes in the
values of these movement-related perceptions.

This forces the discussion back to the level of events instead of continuous
relationships.

I have to admit to being a bit mystified as I read this and what followed,
Bill, because it has little to do with my suggestion. You said:

This picture becomes confused when we take an animal that behaves
essentially continuously and put it into a situation where the _environment_
is set up to create events.

I don't think that the picture becomes confused at all: it is the same rat,
perceiving and behaving as it always does. After the description you
provided of continuous perceptual and behavioral variation as experienced by
the rat and of the artificiality of the division of these continuous changes
into the switch-closure events recorded and programmed by the apparatus, I
was left wondering where you were going with this, whether you had suddenly
decided to eliminate the sequence (or even transition) level from the HPCT
hierarchy of perceptions. You seem to be arguing that these are not
perceptions of the animal, but are artifacts of our recording devices; that
I am being led down the garden path into thinking that what I am seeing in
the operant chamber, with its switch-closure and pellet-delivery events, is
at all representative of the world as the rat sees it.

But my thinking concerning what the rat is learning in the operant chamber
is not predicated at all on the fact that lever-presses are recorded by the
apparatus as switch-closure events, nor that pellet-delivery there is
triggered by that event. I do assume that the rat is able to perceive
changes, particularly sudden ones such as the arrival of a food pellet (or a
predator), and relate these changes to what it was doing at the time
(perceiving relationships). Such coupled changes occur all the time in the
lives of animals, and I would be astounded if animals as sophisticated as
rats were found to be unable to perceive such coincidences and make use of
them in guiding their activities. The question I am grappling with (not
very successfully at present, I'm afraid) is how; my intuition tells me that
there is much more going on in this situation than e-coli type reorganization.

When is behavior? Behavior is always, an ongoing stream, just as
perceptions are. This does not imply that the organism is incapable of
remembering what it was doing when the state of some variable it was
attempting to control suddenly changed for the better. The continuousness
of behavior and perception are not what is at issue; what is at issue is
what is being perceived and how those perceptions may alter subsequent behavior.

Regards,

Bruce

From Tracy Harms (961129.1500 MST)

Regarding the exchange between Bill Powers (961128.2200 MST)
and Bruce Abbott (961127.2010 EST):

My impression is that some pet food, perhaps the pellets you write of, is
created by extrusion and chopping. If so, what if you obtained food rods,
i.e. what you'd have if it hadn't been chopped into pellets. If such food
rods were set to be slid into the cage through a hole by means of a worm
gear connected to, say, an excercize wheel, you would have a continuous
system for delivering food in response to a selected continuous behavior.
Would doing this change the nature of your experiments (re. "events")? If
so, how? If not, why not?

Tracy Bruce Harms tbh@tesser.com
Boulder, Colorado

···

==========================================================================
   "There is only one way I can see for fallible, ignorant human beings
   to live in accord with their own real natures and that is to discard
   forever the principle of controlling each other's behavior, dropping
   even the *desire* to control other people, and seeing at every level
   the fallacy in the logic that leads to such a desire."
                                                   William T. Powers

[

From Bill Powers (961130.0530 MST)]

Bruce Abbott (961129.1440 EST) --

I have to admit to being a bit mystified as I read this and what followed,
Bill, because it has little to do with my suggestion. You said:

This picture becomes confused when we take an animal that behaves
essentially continuously and put it into a situation where the

_environment_ is set up to create events.

I don't think that the picture becomes confused at all: it is the same rat,
perceiving and behaving as it always does. After the description you
provided of continuous perceptual and behavioral variation as experienced by
the rat and of the artificiality of the division of these continuous changes
into the switch-closure events recorded and programmed by the apparatus, I
was left wondering where you were going with this, whether you had suddenly
decided to eliminate the sequence (or even transition) level from the HPCT
hierarchy of perceptions. You seem to be arguing that these are not
perceptions of the animal, but are artifacts of our recording devices; that
I am being led down the garden path into thinking that what I am seeing in
the operant chamber, with its switch-closure and pellet-delivery events, is
at all representative of the world as the rat sees it.

We're talking specifically about the event level, the level at which the
collection of configurations of sensations going through sets of transitions
is recognized as a single event, and corrected if it's not right. I don't
have much doubt that human beings perceive and control events, but
demonstrating that an animal does would require doing experiments
specifically designed to test this idea. You'd have to present the animal
with a "wrong" event and see if the animal changes its behavior so the next
time the event occurs it's the "right" one. The fact that YOU can see events
happening is no indication that the animal can.

But my thinking concerning what the rat is learning in the operant chamber
is not predicated at all on the fact that lever-presses are recorded by the
apparatus as switch-closure events, nor that pellet-delivery there is
triggered by that event. I do assume that the rat is able to perceive
changes, particularly sudden ones such as the arrival of a food pellet (or a
predator), and relate these changes to what it was doing at the time
(perceiving relationships).

Not what it was _doing_ at the time; what it was _perceiving_ at the time.
If the rat was perceiving forces being applied to the lever at the time it
heard and saw and smelled the pellet tumbling into the cup, then what it has
to learn is how to create the same perceptions again, from any starting body
position, orientation, and velocity. That's what learning a transfer
function means. I think it's fairly obvious from watching the rats, however,
that the rats don't initially have any idea of which part of the whole
perceptual world has to be recreated to make the food appear. Was it the
feeling of rearing up on the hind legs? The appearance of the top corner of
the cage in the center of vision of the left eye? The feeling of its fur
brushing against the wall of the cage? A smell in its nose? The sensation of
turning left? The feeling of the paws touching something cold? It's easy for
us to run through such lists because we have language to do it in, and logic
and other useful things that help us sort out the world. But the rat has to
do this without language, and I presume without any systematic kind of
logical symbolic thinking. It's obvious to me how _I_ would solve the
problem of getting the food, but it's not at all obvious how the rat does it.

Such coupled changes occur all the time in the
lives of animals, and I would be astounded if animals as sophisticated as
rats were found to be unable to perceive such coincidences and make use of
them in guiding their activities.

Perceptions do not guide activities; activities control perceptions. The
problem is, as you say, to perceive coincidences (the right ones); but once
they are perceived, the problem becomes that of creating _the same
coincidence_ again. By coincidence, I trust you mean systematic rather than
fortuitous relationships.

When you say rats are "sophisticated" I have to ask, "Compared to whom?" I
think rats are pretty stupid, compared to a person.

The question I am grappling with (not
very successfully at present, I'm afraid) is how; my intuition tells me that
there is much more going on in this situation than e-coli type >reorganization.

When is behavior? Behavior is always, an ongoing stream, just as
perceptions are. This does not imply that the organism is incapable of
remembering what it was doing when the state of some variable it was
attempting to control suddenly changed for the better. The continuousness
of behavior and perception are not what is at issue; what is at issue is
what is being perceived and how those perceptions may alter subsequent
behavior.

The perceptions don't alter the behavior; the behavior alters the
perceptions. The organism doesn't remember what it was doing, it remembers
(if it has memory apparatus to do this with) what it was perceiving. And
then it has to find out how to produce that same perception again, if it
doesn't already have a control system for that perception. There's no _a
priori_ way of altering a perception that hasn't been controlled before; all
you can do is try actions at random until you find one that lessens the
error, and keep altering your organization until your actions reliably
affect the perception and allow you to control it. This, of course, isn't
something you can do consciously; this "you" means "some process inside of you."

The problem with thinking "intuitively" about this, Bruce, is that you have
spent 20 or 30 years training your intuition to think like a behaviorist,
and only about 2 years training it to think like a control theorist. I can
fully appreciate that it would be hard for a behaviorist to see that the
overt actions we see an animal performing are not the most important thing
about its behavior -- that is, it would be hard if no alternative theory had
been encountered.

The basic problem here has to do with system concepts. Behaviorism, as an
organized set of principles and procedures, is a system concept about
organisms. When a PCTer (I mean me, but perhaps others too) sees an
organism, he sees, or imagines, an active entity trying to control its
perceptions relative to internal criteria that define which are bad and
which are good, learning to manipulate its world to be more like the world
it wants to experience. What does the behaviorist who has never heard of PCT
see? I think it's something very different. Of course you're the authority
on what that vision is; I'm not, as I no longer remember what it was like to
think that way, if I ever knew. All I know is that this system concept leads
the behaviorist to interpret stimuli, contingencies, and reinforcers as
governing the organism's actions, changing the organism so its actions
consistently produce the reinforcers. It leads the behaviorist to describe
the organism's actions as "responses" even when the "stimulus" for each
"response" is unknown. It attributes affects to the environment described
variously as "maintaining" or "producing" or "strengthening" or
"extinguishing" or "modifying" the behavior of organisms, with the
environment being spoken of in the active voice and the organism in the
passive voice. The behaviorist clearly sees the locus of control as being in
the environment, where the PCTer sees it as being in the organism. The
behavior of an organism _is controlled by_ its consequences, instead of the
consequences being controlled by the organism's behavior.

This is the source of the tension between us, Bruce. While you're very quick
to grasp the principles of control and the logical procedures involved in
modeling it, and while you have learned to use the language and categories
of control theory competently, you still seem to be operating from the
behaviorist's system concept, so you feel comfortable in saying that the
occurrance of reinforcers "maintains" or "sustains" behavior. You've been
arguing seriously that this is a completely proper description. But it's
proper only under the behavioristic system concept. A control theorist would
never say that; that way of putting it would seem, instantly, backward. It
is the organism that is maintaining or sustaining the reinforcers, which in
fact are not reinforcers but simply sources of perceptions that the organism
wishes, as the time, to maintain in its perceptual world. It's the desire to
maintain those experiences, when possible, that accounts for the actions
that produce the reinforcers. The reinforcers by themselves do nothing.

I don't think you can straddle these system concepts. They're fundamentally
incompatible, even though in some details they can be made to fit together.
As is obvious from your correspondence with me and Rick, the behaviorist
system concept results in clashes with anyone who has fully adopted the PCT
system concept. It can't be helped; your language, your images, your
associations, all have to be maintained in a state consistent with the
overriding system concept. It doesn't feel right to speak in any other way;
if you use a different language, you always have to be translating its terms
into those that fit with the system concept that governs everything. I have
seen my first book cited, by a behaviorist, as "Perception: the control of
behavior." (You can tell where people are coming from by their mistakes; I
have seen quite a few cyberneticists citing Wiener's book as "Cybernetics:
Communication and control in the animal and the machine.")

Best,

Bill P.

[From Bruce Abbott (961130.1055 EST)]

Durango made the weather-news yesterday -- they say 11 inches of snow fell.

Bill Powers (961130.0530 MST) --

Bruce Abbott (961129.1440 EST)

We're talking specifically about the event level, the level at which the
collection of configurations of sensations going through sets of transitions
is recognized as a single event, and corrected if it's not right. I don't
have much doubt that human beings perceive and control events, but
demonstrating that an animal does would require doing experiments
specifically designed to test this idea. You'd have to present the animal
with a "wrong" event and see if the animal changes its behavior so the next
time the event occurs it's the "right" one.

It's been done; they can do it.

But my thinking concerning what the rat is learning in the operant chamber
is not predicated at all on the fact that lever-presses are recorded by the
apparatus as switch-closure events, nor that pellet-delivery there is
triggered by that event. I do assume that the rat is able to perceive
changes, particularly sudden ones such as the arrival of a food pellet (or a
predator), and relate these changes to what it was doing at the time
(perceiving relationships).

Not what it was _doing_ at the time; what it was _perceiving_ at the time.

Somehow I knew you'd nail me for using _doing_ (although I hoped you
wouldn't). Bill, you have used _doing_ this way yourself. By it I
specifically do _not_ mean the actions by which the _doing_ is accomplished.
Doing is controlling perceptions via actions; what has to be recreated via
actions are certain perceptions that existed at the time of pellet
delivery. The problem for the rat is that it doesn't know which of those
perceptions need to be controlled (reestablished) to make the food pellet
appear; indeed, initially it doesn't know whether any of those perceptions
had anything to do with it.

If the rat was perceiving forces being applied to the lever at the time it
heard and saw and smelled the pellet tumbling into the cup, then what it has
to learn is how to create the same perceptions again, from any starting body
position, orientation, and velocity. That's what learning a transfer
function means.

Yes, I understand that. But it _already_ knows how to create the same
perceptions again, except for one: making the pellet appear. What it needs
to learn is which perceptions to create in order to make that happen.

I think it's fairly obvious from watching the rats, however,
that the rats don't initially have any idea of which part of the whole
perceptual world has to be recreated to make the food appear. Was it the
feeling of rearing up on the hind legs? The appearance of the top corner of
the cage in the center of vision of the left eye? The feeling of its fur
brushing against the wall of the cage? A smell in its nose? The sensation of
turning left? The feeling of the paws touching something cold? It's easy for
us to run through such lists because we have language to do it in, and logic
and other useful things that help us sort out the world. But the rat has to
do this without language, and I presume without any systematic kind of
logical symbolic thinking. It's obvious to me how _I_ would solve the
problem of getting the food, but it's not at all obvious how the rat does it.

O.K., so far we agree. You've done a nice job of stating the problem I've
been concerning myself about.

Such coupled changes occur all the time in the
lives of animals, and I would be astounded if animals as sophisticated as
rats were found to be unable to perceive such coincidences and make use of
them in guiding their activities.

Perceptions do not guide activities; activities control perceptions. The
problem is, as you say, to perceive coincidences (the right ones); but once
they are perceived, the problem becomes that of creating _the same
coincidence_ again. By coincidence, I trust you mean systematic rather than
fortuitous relationships.

Language again. I was thinking of behavior as the "doing" again
(controlling perceptions), not as actions, but didn't make that clear. Some
perceptions are not controlled; the rat cannot control the perception that
some perceptions (e.g., the consequences of certain of its own actions)
correlate with other perceptions (e.g., the delivery of a food pellet). But
such perceptions, I am assuming, serve as input to a mechanism that chooses
which of the myriad perceptions under the rat's control to recreate. I
suspect that such a mechanism would be as easily affected by fortuitous as
systematic relationships while the number of samples was still low -- not
enough data to discriminate systematic from chance coincidences.

When you say rats are "sophisticated" I have to ask, "Compared to whom?" I
think rats are pretty stupid, compared to a person.

Sophisticated compared to your apparent opinion of their abilities.

When is behavior? Behavior is always, an ongoing stream, just as
perceptions are. This does not imply that the organism is incapable of
remembering what it was doing when the state of some variable it was
attempting to control suddenly changed for the better. The continuousness
of behavior and perception are not what is at issue; what is at issue is
what is being perceived and how those perceptions may alter subsequent
behavior.

The perceptions don't alter the behavior; the behavior alters the
perceptions. The organism doesn't remember what it was doing, it remembers
(if it has memory apparatus to do this with) what it was perceiving.

See above. Perceptions may alter what the animal does (what perceptions it
attempts to control, to bring into being) by serving as input to a mechanism
that decides on the basis of experience what perceptions to try controlling.

There's no _a
priori_ way of altering a perception that hasn't been controlled before; all
you can do is try actions at random until you find one that lessens the
error, and keep altering your organization until your actions reliably
affect the perception and allow you to control it. This, of course, isn't
something you can do consciously; this "you" means "some process inside of
you."

My proposal is aimed toward providing a more efficient mechanism for doing
this than random assertion (through an as yet unspecified mechanism) of
lower-level perception-references. I am speaking of a "process inside" that
may accomplish this.

The problem with thinking "intuitively" about this, Bruce, is that you have
spent 20 or 30 years training your intuition to think like a behaviorist,
and only about 2 years training it to think like a control theorist. I can
fully appreciate that it would be hard for a behaviorist to see that the
overt actions we see an animal performing are not the most important thing
about its behavior -- that is, it would be hard if no alternative theory had
been encountered.

I think you've been mislead by your misinterpretation of what I was saying,
which is my fault for not being clearer. I hope I've succeeded in getting
it across this time. It's really a very modest proposal, which I see as an
addition to the e-coli random-reorganization proposal as opposed to a
replacement for it. Some important details as to how this mechanism would
work remain to be explicated, but I don't see any difficulties with it in
principle. Working out the details will, however, require some experimental
work.

This is the source of the tension between us, Bruce. While you're very quick
to grasp the principles of control and the logical procedures involved in
modeling it, and while you have learned to use the language and categories
of control theory competently, you still seem to be operating from the
behaviorist's system concept, so you feel comfortable in saying that the
occurrance of reinforcers "maintains" or "sustains" behavior. You've been
arguing seriously that this is a completely proper description. But it's
proper only under the behavioristic system concept. A control theorist would
never say that; that way of putting it would seem, instantly, backward. It
is the organism that is maintaining or sustaining the reinforcers, which in
fact are not reinforcers but simply sources of perceptions that the organism
wishes, as the time, to maintain in its perceptual world. It's the desire to
maintain those experiences, when possible, that accounts for the actions
that produce the reinforcers. The reinforcers by themselves do nothing.

The basic problem there, as I view it, is that I have been unable to get
across to you the fundamental distinction to be made between these two ways
of describing and understanding behavior. The result is that you insist on
applying the vocabulary of mechanism to statements written in the vocabulary
of function, in which terms like "sustain" and "cause" have very different
meanings. The result can only be confusion, as in the last sentence in the
above-quoted paragraph, which would have EAB types sadly shaking their heads
that anyone would think that they believe otherwise.

Be that as it may, my proposal is not founded on an inability on my part to
think like a control theorist, as you suggest. While requiring additional
work to sketch in the proposed mechanism so that the system can be modeled,
I believe there is nothing in it that is inconsistent with mechanistic
principles. Whether it actually applies to any real organisms is an
empirical question.

Regards,

Bruce

From Tracy Harms (1996;11,30.21)

Bill Powers (961130.0530 MST)]

To Bruce Abbott you comment

When you say rats are "sophisticated" I have to ask, "Compared to whom?"
I think rats are pretty stupid, compared to a person.

In addition to Bruce's challenge to your evaluation, I think it is
reasonable to gauge their sophistication relative to most animals. By any
measure of "most" which comes to my mind, they're pretty close to the top
of the pile.

Even when we're talking straight comparison to a person, sometimes rats
look impressive enough. I recently read a war story from a Korean War vet
about the time he found himself, along with several suddenly-deceased
buddies, buried in a bunker by an artillery shell. Also in the same space
was a huge Asian rat. (Sorry I can't locate the story to directly quote
from.) He writes of seeing the rat digging relentlessly at high speed, and
that as the rat seemed to know what it was doing, he decided to follow it.
The two of them dug for quite a distance in the same tunnel to effect their
escape.

Tracy Harms

[

From Bill Powers (961201.0130 MST)]

Bruce Abbott (961130.1055 EST)--

Durango made the weather-news yesterday -- they say 11 inches of snow fell.

Yep. Our surroundings have gone back to "Winter postcard" mode.

We're talking specifically about the event level, the level at which the
collection of configurations of sensations going through sets of
transition is recognized as a single event, and corrected if it's not
right.

It's been done; they can do it.

WHAT's been done? They can do WHAT? "Been there, done that" doesn't tell me
anything. I would really like to know how "they" established that a rat can
control an event-level perception. I'd like to have a nice way of applying
the Test to see if events really can be perceived and controlled by a rat
(or a human being). One way I've thought of is to have a person use a stick
to vary the sound of a phoneme in the middle of a word, a spoken word being
an example of an event. If a distubance is added to the effect of the
joystick, the person would have to move the joystick to make a
continually-repeated word be perceived as the target word. But I've never
tried it. Are you telling me that something similar has been done with rats?

But my thinking concerning what the rat is learning in the operant chamber
is not predicated at all on the fact that lever-presses are recorded by

the apparatus as switch-closure events, nor that pellet-delivery there is

triggered by that event. I do assume that the rat is able to perceive
changes, particularly sudden ones such as the arrival of a food pellet

(or >>>a predator), and relate these changes to what it was doing at the time

(perceiving relationships).

Not what it was _doing_ at the time; what it was _perceiving_ at the time.

Somehow I knew you'd nail me for using _doing_ (although I hoped you
wouldn't). Bill, you have used _doing_ this way yourself.

I was trying to point out that what you're describing is control of a
relationship between two sets of perceptions. The perceptions in question
are (1) the arrival of food, and (2) the remainder of the visual, olfactory,
kinesthetic and so on perceptions that were going on just prior to the
arrival of food. The "suddenness" of the perceptual food-event is
irrelevant. The rat might also learn to control relationships involving
steady perceptions, such as the color of a stimulus light that is on while a
particular environmental condition persists, or the position of a
food-bearing tray that a rat can pull toward itself (like the suggestion
that the rat' actions cause a food stick to extrude into the cage -- was
that Tracy? Chris?).

The reason that the rats fumble around at the bar for so many hours is that
they DO NOT initially have a control system for controlling the perceptions
relevant to depressing the bar in the right way. They don't know which
perceptions are relevant, and the process by which they find out, and learn
to control them, is apparently one of trial and error. It's interesting that
they learn rather quickly the approximate place in which to fumble around
(although it wouldn't take a human being that long). Finding the right
location in which to behave happens fairly efficiently, perhaps more
efficiently than the E. coli model would suggest (although we won't know
that until we try a model). Figuring out which effects of behaving in that
location are related to the appearance of the food takes longer, and looks
to me a lot like the E. coli style of reorganization.

... I
specifically do _not_ mean the actions by which the _doing_ is accomplished.
Doing is controlling perceptions via actions; what has to be recreated via
actions are certain perceptions that existed at the time of pellet
delivery. The problem for the rat is that it doesn't know which of those
perceptions need to be controlled (reestablished) to make the food pellet
appear; indeed, initially it doesn't know whether any of those perceptions
had anything to do with it.

I agree. And initially it may not even know how to recreate any of those
perceptions such as the feeling of the lever bottoming out.

If the rat was perceiving forces being applied to the lever at the time it
heard and saw and smelled the pellet tumbling into the cup, then what it
has to learn is how to create the same perceptions again, from any

starting >>body position, orientation, and velocity.

Yes, I understand that. But it _already_ knows how to create the same
perceptions again, except for one: making the pellet appear.

I dispute that. Why does it take the rats so many sessions to learn to make
the lever go down?

What it needs
to learn is which perceptions to create in order to make that happen.

That's just the part at the relationship level. The videos I saw of our rats
in their first several dozen sessions don't convince me that the rats even
knew how to create the relevant perceptions.

I think it's fairly obvious from watching the rats, however,
that the rats don't initially have any idea of which part of the whole
perceptual world has to be recreated to make the food appear. Was it the
feeling of rearing up on the hind legs? The appearance of the top corner of
the cage in the center of vision of the left eye? The feeling of its fur
brushing against the wall of the cage? A smell in its nose? The sensation
of turning left? The feeling of the paws touching something cold?

O.K., so far we agree. You've done a nice job of stating the problem I've
been concerning myself about.

The problem I've stated is that of learning to control a relationship
between two perceptions which may or may not be event-perceptions. If the
events or transitions or configurations are already under control, then
that's the only problem (except for how to convert an error in the
relationship into the specific changes in lower-level reference signals that
will tend to correct the error). If the lower-level systems for controlling
the relevant perceptions don't exist, however, then they, too, have to be
learned. It seems to me that what we saw early in our observations of the
rats was that they had to learn at several levels. A rat "digging" at a
lever without depressing it has obviously not yet figured out how to make
the critical perception occur.

Such coupled changes occur all the time in the
lives of animals, and I would be astounded if animals as sophisticated as
rats were found to be unable to perceive such coincidences and make use of
them in guiding their activities.

Perceptions do not guide activities; activities control perceptions. The
problem is, as you say, to perceive coincidences (the right ones); but once
they are perceived, the problem becomes that of creating _the same
coincidence_ again. By coincidence, I trust you mean systematic rather than
fortuitous relationships.

Language again. I was thinking of behavior as the "doing" again
(controlling perceptions), not as actions, but didn't make that clear.

That's not the problem. You're assuming that as soon as the animal has
identified the critical perception, it already knows how to make it occur.
The perception that the rat already knows how to control -- say, the
"digging" motion at the bar -- is not sufficient to affect the critical
perception, even supposing that the rat has figured out what it is. This
motion control system must be reorganized until the ineffective aspects of
it are changed to become effective. This requires reorganizing output
connections as well as input functions.

When you say rats are "sophisticated" I have to ask, "Compared to whom?" I
think rats are pretty stupid, compared to a person.

Sophisticated compared to your apparent opinion of their abilities.

When I see a rat lightly touching the lever and nosing around it and patting
it for the 2000th time, I want to tell it "Push DOWN, dummy!" But that is
anthropomorphizing; the rat wouldn't know what "down" means, or "push."
What's remarkable about the rat, as Dr. Johnson said of the dog walking on
its hind legs, isn't how well it performs, but the fact that it can perform
at all.

In modeling behavior, one must always ask what is the LEAST complex ability
that would account for observations. Look at the Crowd program. People are
always reading complexity abilities into the behavior they see, which I know
for a fact are not there.

The perceptions don't alter the behavior; the behavior alters the
perceptions. The organism doesn't remember what it was doing, it remembers
(if it has memory apparatus to do this with) what it was perceiving.

See above. Perceptions may alter what the animal does (what perceptions it
attempts to control, to bring into being) by serving as input to a mechanism
that decides on the basis of experience what perceptions to try controlling.

Here we go with that "functional" stuff again. And the word I was thinking
of wasn't "stuff."

... you insist on
applying the vocabulary of mechanism to statements written in the vocabulary
of function, in which terms like "sustain" and "cause" have very different
meanings. The result can only be confusion, as in the last sentence in the
above-quoted paragraph, which would have EAB types sadly shaking their heads
that anyone would think that they believe otherwise.

"The vocabulary of function" is an abomination. I've seen biologists using
it, too: it lets them say that the wasp returns its larvae to the hole in
order to protect them from predators, while simultaneously denying that that
is what they mean. The reason they do this is that they don't know any other
way to say it; it's a way of defusing purposive language even when that
seems the only appropriate description.

What the vocabulary of function does is to let the behaviorist go right on
talking as if the environment controls behavior, while insisting that this
is merely an objective report of observations. This is utter nonsense. It's
just a rationalization. It lets you use causal terms like sustain and alter
and cause, with all their connotations and denotations in other
circumstances, while maintaining innocence about actually intending those
connotations and meanings. It's just like a die-hard alchemist saying "What
I mean by the affinity of one object for another is Newton's Law of
Universal Gravitational Attraction," and then proceeding to explain that
objects fall because they have an affinity for Earth. That's not going to
fool anybody.

Be that as it may, my proposal is not founded on an inability on my part to
think like a control theorist, as you suggest.

I didn't suggest that. I think you can think like a control theorist, when
you're not busy defending the reasoning methods of behaviorists. What I said
was that you're having a hard time discarding the system concept behind
behaviorism, together with the network of principles and procedures (and
methods of reasoning) that go with it.

While requiring additional
work to sketch in the proposed mechanism so that the system can be modeled,
I believe there is nothing in it that is inconsistent with mechanistic
principles. Whether it actually applies to any real organisms is an
empirical question.

I'm all for the modelling. But once you have such a model, what are you
going to do? Are you going to say, "SEE? The reinforcements REALLY DO
sustain the pressing"? You can apply the same abominable functional
"analysis" to the model's behavior that you apply to the real behavior, if
you're still convinced that you can see such a causal connection. The model
won't settle this question. Only mathematical (not "mechanistic") rigor will
settle it.

Best,

Bill P.